Qwen3-ASR（原生 Transformers）

概述

Qwen3-ASR 系列包含 Qwen3-ASR-1.7B 和 Qwen3-ASR-0.6B，支持 52 种语言和方言的语种识别与 ASR。两者均利用大规模语音训练数据及其基础模型 Qwen3-Omni 强大的音频理解能力。1.7B 版本在开源 ASR 模型中达到了 SOTA 性能，并与最强的专有商业 API 具有竞争力。

主要特性：

• 一体化： 支持 30 种语言和 22 种中文方言的语种识别与语音识别，包括多个国家和地区的英语口音。
• 优秀且快速： 在复杂声学环境下实现高质量、鲁棒的识别。Qwen3-ASR-0.6B 在 128 并发下吞吐量达到 2000×。两个模型均支持单模型流式/离线统一推理，并处理长音频。
• 强制对齐： Qwen3-ForcedAligner-0.6B 支持对最长 5 分钟语音中任意单元（11 种语言）的时间戳预测，在准确率上超越基于 E2E 的强制对齐模型。

模型架构

可用检查点

模型	支持语言	支持方言	推理模式	音频类型
Qwen/Qwen3-ASR-1.7B-hf 和 Qwen/Qwen3-ASR-0.6B-hf	中文 (zh)、英语 (en)、粤语 (yue)、阿拉伯语 (ar)、德语 (de)、法语 (fr)、西班牙语 (es)、葡萄牙语 (pt)、印尼语 (id)、意大利语 (it)、韩语 (ko)、俄语 (ru)、泰语 (th)、越南语 (vi)、日语 (ja)、土耳其语 (tr)、印地语 (hi)、马来语 (ms)、荷兰语 (nl)、瑞典语 (sv)、丹麦语 (da)、芬兰语 (fi)、波兰语 (pl)、捷克语 (cs)、菲律宾语 (fil)、波斯语 (fa)、希腊语 (el)、匈牙利语 (hu)、马其顿语 (mk)、罗马尼亚语 (ro)	安徽、东北、福建、甘肃、贵州、河北、河南、湖北、湖南、江西、宁夏、山东、陕西、山西、四川、天津、云南、浙江、粤语（香港）、粤语（广东）、吴语、闽南语	离线 / 流式	语音、歌声、带背景音乐歌曲
Qwen/Qwen3-ForcedAligner-0.6B-hf	中文、英语、粤语、法语、德语、意大利语、日语、韩语、葡萄牙语、俄语、西班牙语	—	NAR	语音

---

使用

Qwen3-ASR 在 🤗 Transformers 中原生支持。在它成为官方 Transformers 版本的一部分之前，请从源码安装：

pip install git+https://github.com/huggingface/transformers

简单转录

apply_transcription_request 为你处理 chat-template 格式化，是推荐的入口点。

from transformers import AutoProcessor, AutoModelForMultimodalLM

model_id = "Qwen/Qwen3-ASR-0.6B-hf"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForMultimodalLM.from_pretrained(model_id, device_map="auto")
print(f"Model loaded on {model.device} with dtype {model.dtype}")

inputs = processor.apply_transcription_request(
    audio="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-ASR-Repo/asr_en.wav",
).to(model.device, model.dtype)

output_ids = model.generate(**inputs, max_new_tokens=256)
generated_ids = output_ids[:, inputs["input_ids"].shape[1]:]

# Raw output includes language tag and <asr_text> marker
raw = processor.decode(generated_ids)[0]
print(f"Raw: {raw}")

# Parsed output: dict with "language" and "transcription"
parsed = processor.decode(generated_ids, return_format="parsed")[0]
print(f"Parsed: {parsed}")

# Extract only the transcription text
transcription = processor.decode(generated_ids, return_format="transcription_only")[0]
print(f"Transcription: {transcription}")

"""
Raw: language English<asr_text>Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.
Parsed: {'language': 'English', 'transcription': 'Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.'}
Transcription: Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.
"""

语言提示

传入语言提示以跳过自动检测。

from transformers import AutoProcessor, AutoModelForMultimodalLM

model_id = "Qwen/Qwen3-ASR-0.6B-hf"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForMultimodalLM.from_pretrained(model_id, device_map="auto")

# Without language hint (auto-detect)
inputs = processor.apply_transcription_request(
    audio="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-ASR-Repo/asr_zh.wav",
).to(model.device, model.dtype)
output_ids = model.generate(**inputs, max_new_tokens=256)
generated_ids = output_ids[:, inputs["input_ids"].shape[1]:]
print(f"Auto-detect: {processor.decode(generated_ids, return_format='transcription_only')[0]}")

# With language hint (language code or full name both accepted)
inputs = processor.apply_transcription_request(
    audio="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-ASR-Repo/asr_zh.wav",
    language="Chinese",  # or "zh"
).to(model.device, model.dtype)
output_ids = model.generate(**inputs, max_new_tokens=256)
generated_ids = output_ids[:, inputs["input_ids"].shape[1]:]
print(f"With hint:   {processor.decode(generated_ids, return_format='transcription_only')[0]}")

批量推理

传入音频路径列表和可选语言，一次调用转录多个文件。

from transformers import AutoProcessor, AutoModelForMultimodalLM

model_id = "Qwen/Qwen3-ASR-0.6B-hf"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForMultimodalLM.from_pretrained(model_id, device_map="auto")

audio = [
    "https://huggingface.co/datasets/bezzam/audio_samples/resolve/main/librispeech_mr_quilter.wav",
    "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-ASR-Repo/asr_zh.wav",
]

inputs = processor.apply_transcription_request(
    audio, language=[None, "zh"],
).to(model.device, model.dtype)

output_ids = model.generate(**inputs, max_new_tokens=256)
generated_ids = output_ids[:, inputs["input_ids"].shape[1]:]
transcriptions = processor.decode(generated_ids, return_format="transcription_only")

for i, text in enumerate(transcriptions):
    print(f"Audio {i + 1}: {text}")

Chat 模板

apply_transcription_request 是 apply_chat_template 的便捷封装。直接使用 chat 模板可以获得更多控制，例如通过系统消息提供语言提示。

from transformers import AutoProcessor, Qwen3ASRForConditionalGeneration

model_id = "Qwen/Qwen3-ASR-0.6B-hf"
processor = AutoProcessor.from_pretrained(model_id)
model = Qwen3ASRForConditionalGeneration.from_pretrained(model_id, device_map="auto")

chat_template = [
    [
        {"role": "system", "content": [{"type": "text", "text": "English"}]},
        {
            "role": "user",
            "content": [
                {
                    "type": "audio",
                    "path": "https://huggingface.co/datasets/bezzam/audio_samples/resolve/main/librispeech_mr_quilter.wav",
                },
            ],
        },
    ],
    [
        {
            "role": "user",
            "content": [
                {
                    "type": "audio",
                    "path": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-ASR-Repo/asr_zh.wav",
                },
            ],
        },
    ],
]

inputs = processor.apply_chat_template(
    chat_template, tokenize=True, return_dict=True,
).to(model.device, model.dtype)

output_ids = model.generate(**inputs, max_new_tokens=256)
generated_ids = output_ids[:, inputs["input_ids"].shape[1]:]
transcriptions = processor.decode(generated_ids, return_format="transcription_only")
for text in transcriptions:
    print(text)

训练 / 微调

from transformers import AutoProcessor, Qwen3ASRForConditionalGeneration

model_id = "Qwen/Qwen3-ASR-0.6B-hf"
processor = AutoProcessor.from_pretrained(model_id)
model = Qwen3ASRForConditionalGeneration.from_pretrained(model_id, device_map="auto")
model.train()

chat_template = [
    [
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.",
                },
                {
                    "type": "audio",
                    "path": "https://huggingface.co/datasets/bezzam/audio_samples/resolve/main/librispeech_mr_quilter.wav",
                },
            ],
        }
    ],
]

inputs = processor.apply_chat_template(
    chat_template, tokenize=True, return_dict=True, output_labels=True,
).to(model.device, model.dtype)

loss = model(**inputs).loss
print("Loss:", loss.item())
loss.backward()

强制对齐（词级时间戳）

使用 Qwen3ASRForTokenClassification 从转录文本获取词级时间戳。先用 ASR 模型转录，然后用强制对齐器进行对齐。

支持的语言：中文、英语、粤语、法语、德语、意大利语、日语、韩语、葡萄牙语、俄语、西班牙语。

日语需要 nagisa，韩语需要 soynlp：pip install nagisa soynlp

import torch
from transformers import AutoProcessor, AutoModelForMultimodalLM, AutoModelForTokenClassification

asr_model_id = "Qwen/Qwen3-ASR-0.6B-hf"
aligner_model_id = "Qwen/Qwen3-ForcedAligner-0.6B-hf"

asr_processor = AutoProcessor.from_pretrained(asr_model_id)
asr_model = AutoModelForMultimodalLM.from_pretrained(asr_model_id, device_map="auto")

aligner_processor = AutoProcessor.from_pretrained(aligner_model_id)
aligner_model = AutoModelForTokenClassification.from_pretrained(
    aligner_model_id, dtype=torch.bfloat16, device_map="auto"
)

audio_url = "https://huggingface.co/datasets/bezzam/audio_samples/resolve/main/librispeech_mr_quilter.wav"

# Step 1: Transcribe
inputs = asr_processor.apply_transcription_request(audio=audio_url)
inputs = inputs.to(asr_model.device, asr_model.dtype)
output_ids = asr_model.generate(**inputs, max_new_tokens=256)
generated_ids = output_ids[:, inputs["input_ids"].shape[1]:]
parsed = asr_processor.decode(generated_ids, return_format="parsed")[0]
transcript = parsed["transcription"]
language = parsed["language"] or "English"

# Step 2: Prepare alignment inputs
aligner_inputs, word_lists = aligner_processor.prepare_forced_aligner_inputs(
    audio=audio_url, transcript=transcript, language=language,
)
aligner_inputs = aligner_inputs.to(aligner_model.device, aligner_model.dtype)

# Step 3: Run forced aligner
with torch.inference_mode():
    outputs = aligner_model(**aligner_inputs)

# Step 4: Decode timestamps
timestamps = aligner_processor.decode_forced_alignment(
    logits=outputs.logits,
    input_ids=aligner_inputs["input_ids"],
    word_lists=word_lists,
    timestamp_token_id=aligner_model.config.timestamp_token_id,
)[0]

for item in timestamps:
    print(f"{item['text']:<20} {item['start_time']:>8.3f}s → {item['end_time']:>8.3f}s")

"""
Word                  Start (s)    End (s)
------------------------------------------
Mr                        0.560      0.800
Quilter                   0.800      1.280
is                        1.280      1.440
the                       1.440      1.520
apostle                   1.520      2.080
...
"""

Pipeline 使用

from transformers import pipeline

model_id = "Qwen/Qwen3-ASR-0.6B-hf"
pipe = pipeline("any-to-any", model=model_id, device_map="auto")

chat_template = [
    {
        "role": "user",
        "content": [
            {
                "type": "audio",
                "path": "https://huggingface.co/datasets/bezzam/audio_samples/resolve/main/librispeech_mr_quilter.wav",
            },
        ],
    }
]
outputs = pipe(text=chat_template, return_full_text=False)
raw_text = outputs[0]["generated_text"]

# Use processor helper to extract transcription
transcription = pipe.processor.extract_transcription(raw_text)
print(f"Transcription: {transcription}")

---

速度与内存改进

Torch compile

ASR 和强制对齐模型均支持 torch.compile。强制对齐器尤其适合，因为它只执行一次前向传播，无需自回归解码，非常适合批量时间戳处理工作流。

在 A100 上，我们观察到强制对齐器加速约 2.5 倍，ASR 在 batch size 为 4 时 generate 加速约 2.4 倍。

import torch
from transformers import AutoProcessor, AutoModelForMultimodalLM

model_id = "Qwen/Qwen3-ASR-0.6B-hf"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForMultimodalLM.from_pretrained(model_id, dtype=torch.bfloat16).to("cuda").eval()

audio_url = "https://huggingface.co/datasets/bezzam/audio_samples/resolve/main/librispeech_mr_quilter.wav"
inputs = processor.apply_transcription_request(
    audio=[audio_url] * 4,
).to("cuda", torch.bfloat16)

model.forward = torch.compile(model.forward)

# Warmup
with torch.inference_mode():
    for _ in range(3):
        _ = model.generate(**inputs, max_new_tokens=256, do_sample=False)

# Inference
with torch.inference_mode():
    output_ids = model.generate(**inputs, max_new_tokens=256, do_sample=False)

generated_ids = output_ids[:, inputs["input_ids"].shape[1]:]
print(processor.decode(generated_ids, return_format="transcription_only")[0])

---

评估

HuggingFace Open ASR Leaderboard 上的 WER（2026 年 6 月 26 日）：

模型	平均 WER	AMI	Earnings22	GigaSpeech	LS Clean	LS Other	SPGISpeech	VoxPopuli
Qwen3-ASR-1.7B-hf	5.59	9.26	9.88	7.25	1.24	2.92	2.58	5.99
Qwen3-ASR-0.6B-hf	6.31	10.57	10.72	7.65	1.69	3.97	2.74	6.80

---

引用

@article{Qwen3-ASR,
  title={Qwen3-ASR Technical Report},
  author={Xian Shi, Xiong Wang, Zhifang Guo, Yongqi Wang, Pei Zhang, Xinyu Zhang, Zishan Guo,
          Hongkun Hao, Yu Xi, Baosong Yang, Jin Xu, Jingren Zhou, Junyang Lin},
  journal={arXiv preprint arXiv:2601.21337},
  year={2026}
}